Anthropic is testing a Model Hardware Standard that gives AI agents a common way to discover, read and control programmable equipment. The research preview extends the idea behind Model Context Protocol from software tools to microscopes, robotic arms, liquid handlers and other physical devices.
Each machine needs a reusable driver describing its commands, measurements and physical limits. Anthropic says this can reduce multi-device integration from weeks or months to hours or minutes. At Carnegie Mellon, a team connected incompatible lab equipment across three computers in about eight hours. At QuEra, a script developed through the system restored a laser after disruptions in 695 of 700 blind-test attempts, with the language model removed from the final control loop.
The experiments also exposed the limit of text-based reasoning about physical systems. During a Genentech protein assay, Claude repeatedly adjusted software parameters after bubbles in a viscous sample caused errors; a person had to explain the physical cause before it found a workable correction. The partner results have not been independently verified. Anthropic developed the standard with HHMI Janelia Research Campus and plans an open-source release after the initial preview. A shared interface may make automation easier to reuse, but laboratories still need device-specific safety limits, human oversight and conventional scripts for validated production runs.